--> From Unconventional Reservoir Characterization, 3D Seismic multi-attribute analysis and machine learning guided geocellular modeling to Well Performance (EUR) Simulation: Woodford Shale Case Study in North of Oklahoma, USA
[First Hit]

2019 AAPG Annual Convention and Exhibition:

Datapages, Inc.Print this page

From Unconventional Reservoir Characterization, 3D Seismic Previous HitmultiNext Hit-Previous HitattributeNext Hit Previous HitanalysisNext Hit and machine learning guided geocellular modeling to Well Performance (EUR) Simulation: Woodford Shale Case Study in North of Oklahoma, USA

Abstract

The objective of our study is to use data from every scale for reservoir characterization and eventually reservoir simulation. The current area of study is Woodford Shale in Oklahoma. Due to enormous pay zone and thus oil and gas reserves, the Woodford Shale is the center of attraction for major oil operators in Oklahoma currently. In this study we have used well logs to obtain high vertical resolution petrophysical and geomechanical properties such as: Porosity, total organic carbon (TOC), Young’s Modulus, Shear Modulus, Biot’s coefficient, Poisson’s ratio, stresses etc. We have used seismic inversion derived impedance volumes to populate these properties in the geo-cellular model area to obtain better stratigraphic control on these properties. To obtain the type curves (TC’s), we have used unsupervised machine learning methodologies such as Previous HitSelfNext Hit Previous HitOrganizingNext Hit maps (SOM) and Generative Topographic Mapping (GTM) to cluster the reservoir properties together. From these clusters, we have identified four type curve (TC) areas. To demonstrate the physical significance of these four type curve areas, we have used a fully compositional, fully coupled geomechanical Equation of State (EOS) simulator to simulate the Stimulated Rock Volume (SRV) as a function of stress and eventually production. Our results show, the integration from seismic to production scale provides excellent control on the reservoir characterization and bolsters confidence on the static model. Out of all clustering techniques, SOM works best in our case and is consistent with the regional geology. More number of samples in case of seismic provides better horizontal control on geomechanical properties and hence modeling of the SRV and ultimately predicting production behavior.